What we know

AMD and Intel have announced a collaboration to simplify AI development on x86 CPUs by merging their optimization efforts for PyTorch within the oneDNN library. This joint work aims to provide developers with a unified software path that supports both AMD and Intel CPUs, streamlining AI workloads on these platforms. The companies will combine their x86 CPU optimization work for PyTorch in oneDNN. This collaboration targets a common software path for PyTorch across AMD and Intel processors. The effort focuses on improving AI development efficiency on x86 architectures.

Why it matters

PyTorch is a leading open-source framework for machine learning, widely used by developers and researchers. Previously, AMD and Intel maintained separate optimizations tailored to their own CPUs, which could complicate development and deployment. By unifying their efforts in oneDNN, an open-source performance library for deep learning, the companies aim to reduce fragmentation and offer more consistent performance across x86 CPUs.

What is still unknown

Details on the timeline for release, the range of supported programming languages, and independent evaluations of performance or accuracy have not been disclosed. No independent tests of the combined optimizations’ effectiveness or energy efficiency are available yet.